Abstract:
Outcomes studied in social science and health research frequently take the form of fractions or percentages with a defined lower and upper limit, such as the percentage of medication doses taken, the rate of condom-protected sex, and the number of substance use-related problems endorsed in a screening questionnaire. Such data commonly exhibit floor and ceiling effects due to many participants that never engage in the outcome (e.g., never take prescribed medication) or are consistently at the upper limit (e.g., take all prescribed doses). Prevailing statistical approaches used to analyze such data do not fully account for clusters of responses at the lower and upper limits, which risk invalid conclusions about the effectiveness of interventions and theoretical models. We introduce an accessible extension to zero-inflated regression, the marginalized zero- and N-inflated binomial (MZNIB) model, that can analyze fractions and percentage data on the entire range between zero and 100% with greater accuracy than prevailing approaches when assessing naturally bounded outcomes with floor and ceiling effects.
Dr. David Huh is the Director of the Methods Division at the UW Indigenous Wellness Research Institute and the Associate Director of the Methods Core at the UW Behavioral Research Center for HIV (BIRCH). A key emphasis of Dr. Huh’s research is increasing the accessibility of statistical approaches that can more powerfully and accurately assess behavioral health interventions and test theoretical models of health for both specific groups and broad populations.
